I am just looking here for reasons why I can’t use GPU acceleration in Blender, Cinema 4D and other software with GPU rendering capability, as most other places online have either told me to go here or that they can’t help me. I was in NVIDIA live chat for hours with no success. I run Windows 10 latest with updated graphics drivers.
GPU-Z shows no CUDA capability, and CUDA-Z doesn’t start, stating there is no compatible graphics card.
What I have tried:
-Updated to studio and game ready drivers
-Wiped previous drivers with DDU
-Downgraded drivers to 470
-Installed latest CUDA toolkit with Microsoft Visual Studio Community
-Reseated GPU in primary and secondary PCI slots
What I have not tried:
-Reinstalling windows
-Testing card on other device (I have no other PC to test on)
It is also worth mentioning my old GTX 970 can use CUDA acceleration just fine, but on this card I can’t.
I apologize if there isn’t enough information to determine what is actually going on, but I have run out of places to get answers. I can provide whatever may be needed to troubleshoot should there be logs or information that could be relevant.
Any comments, questions or information is greatly appreciated!
Which version of Windows 10, specifically (e.g. mine is: Microsoft Windows 10 Pro for Workstations)? What NVIDIA driver version, specifically (e.g. mine is: 516.25)? When you use Windows’s System Information app, do you see the RTX 3060 listed under System Summary | Components | Display? When you right-click on the Windows desktop, does the pop-up menu include NVIDIA Control Panel and NVIDIA RTX Desktop Manager?
Does GPU-Z recognize the RTX 3060 as such (Graphics Card tab)? If so, when you switch to the Sensors tab, does it show items such as GPU Clock, GPU Temperature, GPU Load, etc? If GPU-Z recognizes the GPU just fine, but the CUDA box on the Graphics Card tab is not checked, you are likely missing the (correct) CUDA driver. Go to this page to download the latest CUDA software, and install it:
The RTX 3060 requires a relatively recent version of CUDA. If you are trying to use a CUDA version that worked with the much older GTX 970, that could easily be much too old to run an RTX 3060. The TechPowerUp database shows the GTX 970 as a GPU with compute capability 5.2 (Maxwell class). If you have that GPU still in your system with the RTX 3060, it should still work with the most recent CUDA version, but it may be on the list of deprecated hardware.
The most obvious answer: have you validated with the Nvidia SMI tool that compute is enabled on the GPU?
It is possible to turn on/off compute mode with SMI.
Thanks for the reply, I have Windows 10 Home (Build 19044)
Yes, my graphics card displays properly in msinfo and works in every video game as well, giving pretty much the exact performance I expect out of it, nothing noticeably wrong with it there.
As for GPU-Z, everything on the page is completely accurate on all tabs, except for the CUDA box in the graphics card tab, it is unchecked regardless if launched as admin or not.
NVIDIA-control panel is slow, (I’ve heard its slow for everyone lol) but all features work just as expected.
My driver version is game ready driver 516.59, I have already tried switching over to the studio ready driver and downgraded to 470 for game ready and studio (this is the minimum driver version Blender requires for CUDA acceleration) but nothing had an effect. And as stated above I did use DDU to wipe old drives when installing the new ones, it was a very slow process but I did it all properly just in case.
I have already installed the latest version of the CUDA toolkit, along with Microsoft visual studio, I even reinstalled it and it still doesn’t solve the problem.
As I am aware, this is a command used in cmd. Upon typing “nvidia-smi” into cmd I get this reply regardless if launched as admin or not.
I apologize if I am incorrect, but from what I understand, the displayed “CUDA Version - 11.7” is merely saying that 11.7 is the version capable of running on the card, not necessarily that it actually is running. How can I actually change settings through this command?
Just to confirm: You first installed MSVS (2017, or 2019, or 2022), then installed CUDA 11.7, correct? When installing CUDA, you chose the default installation, and it ran to completion successfully, without reporting any problems?
Yes, I have installed Visual Studio Community 2022, then installed the latest CUDA toolkit without any problems. Going to try updating BIOS soon if nvidia-smi commands end up doing nothing, if it gets solved I will leave detailed instructions on how it was done.
According to the screen dump, the compute mode in your GPU is set to “Default”, which means that CUDA does not seem to be disabled and nvidia-smi is not going to fix things.
If you’re interested in what it can do, nvidia-smi -h prints out usage information.
The unchecked CUDA box in GPU-Z may be a red herring. It only tells us whether GPU-Z can (by unknown means) detect the presence of CUDA. There may be a bug or incompatibility making the detection mechanism fail. Are you running the latest version of GPU-Z? Version 2.47.0 seems to be the latest at this time.
CUDA ships with a pre-compiled program deviceQuery.exe. What happens when you try to run that?
Can you successfully build any of the simplest of the sample apps that come with CUDA? Since there could be an issue with CUDA integration into MSVS (such as faulty solution files; I have vague memories that this has happened before when CUDA versions changed), try building by running nvcc from the Command Prompt. Ypu may need to manually supply some include file and/or library locations when you do that, but it is definitely possible. I never use an IDE when programming with CUDA.
Gpu-z is very good, e.g. it once showed it cannot see PhysX, even though it was installed and all, only deleting and installing again helped. It did not work indeed, which was not obvious.
Try sigverif.exe, it will show all not signed files including not signed due to their absence.